AI at the heart of z/OS 3.1
Today marks the beginning of a new era for IBM z/OS, the operating system that has been around for more than 50 years of continuous innovation, supporting business transformation, and addressing businesses’ needs to deliver performance, security, and resiliency for mission-critical workloads.
To continue meeting future challenges while reducing the z/OS skills requirements, IBM z/OS is now ready to embrace AI.
The new z/OS 3.1 version delivers an AI framework designed to support AI capabilities inside z/OS and its components that enable system self-management to optimise IT processes, improve performance, and prevent problems.
The initial application for AI in z/OS is now available within the AI-powered WLM batch initiator management. You can use the z/OS Workload Manager component incorporated with AI to predict upcoming batch workloads and react automatically and proactively for optimized system resources.
This blog describes the why and how behind the AI infusion into z/OS and provides you with an idea about what you need to get started.
Why AI infusion into z/OS?
Today, IBM Z organisations are increasingly being confronted with the mainframe skills gap challenge.
Even if the mainframe of today is not the same as a couple of decades ago, novice users still need time to get up to speed to perform their tasks more efficiently.
Thanks to continuous modernisation and numerous simplification approaches, novice personas can already leverage more familiar ways to interact with the system. However, when it comes to performing complex tasks and undergoing IT processes, early and mid-tenure users might still require unique skills, experience, and expertise to perform those tasks successfully.
The new AI capabilities build the base to establish an intelligent operating system with self-management ability designed to help novice users perform their tasks efficiently, avoid trial and error approaches, eliminate fine tuning overhead, and increase their productivity.
AI infusion into z/OS – making the impossible possible
Infusing AI capabilities into a target environment requires the development and operationalization of AI and decision models. The operationalization process is a complex phase that includes the integration of the model in the target application or environment with a continuous loop of data collection, experimentation to improve accuracy and performance, evaluation throughout a deployment process, and monitoring of model performance in production. This complex process requires a set of capabilities that focuses on the life cycle management of the model, usually called AI model operationalization (ModelOps) or Machine Learning operationalization (MLOps) platform. In the following, we will use the term MLOps to refer to this sought platform.
What makes our AI infusion story and the requirements for that needed MLOps platform special is the fact that the target environment is not an application compared to typical AI infusion use cases. As a matter of fact, the AI model is operationalised at the heart of the operating system that includes “aged” components which are not Unix System Services (USS) supported and therefore not ready to communicate with the modern world of AI tools.
The good news is that the major challenges have been successfully addressed. The required platform is now available and referred to as the AI Framework for IBM z/OS. It opens opportunities and paves the way to deliver meaningful AI capabilities infused into components in a common and simplified way, while optimizing the time-to-market and time-to-value.
AI Framework for IBM z/OS – what’s in for you?
As mentioned earlier, the AI Framework for IBM z/OS provides a set of capabilities that support the end-to-end AI model life cycle including AI model (re) training, deployment, and management of AI assets as well as data management.
Those capabilities are mostly provided within the new offering called AI System Services for IBM z/OS which you need to order on shopZ. The good news is that when leveraging AI for z/OS base components, this offering is at no charge for you. It represents a bundling of two strategic offerings in the AI and AIOps portfolio, namely, IBM Z Common Data Provider (ZCDP) and Watson Machine Learning for z/OS Core Edition (WMLz Core). This shows that inside IBM’s offerings, we are eating our own soup.
The AI System Services for IBM z/OS integrates seamlessly with the rest of the framework components that are delivered as part of the z/OS base. This includes EzNoSQL, the data store that is intended to collect the training and inference data, allowing you to train the model with your own historical data and the model to perform inference based on the fresh data.
What about the challenge to let the “aged” components talk to AI? To bridge the gap between the z/OS traditional components and the AI world, z/OS is providing a new component called AI Base Component for IBM z/OS that transfers the requests for prediction - also called inference requests - coming from the components to the AI model server (WMLz Core) and provide the response back to the component.
All the capabilities described so far represent the engine under the hood that you do not need or want to care about or interact with, once they are installed and up and running. However, as soon as your AI system is installed and configured, you might want to get on board and start using your Cockpit to drive and control your AI capabilities. This is exactly the moment where you would need the AI control interface for IBM z/OS, a z/OSMF plugin, to select the AI capability, train your model, switch the AI mode to enabled, disabled, or simulating. You can also have a look at views that allow you to double check if the connection to and between your components is OK.
Remember that before you reach the point of onboarding, your team needs to go thru the check-in process and perform some installation and configuration. This process is based on z/OSMF workflows and designed for more simplification to help you speed up, optimize the configuration experience, and start your AI journey on z/OS in time.
AI-powered WLM batch initiator management
AI-powered WLM batch initiator management is the first use case to provide AI capabilities infused in z/OS.
z /OS Workload Manager already helps automate resource allocation for z/OS, However, workloads may follow distinct cycles over time. Furthermore, spikes may impact system performance especially because initiators are started after the workload arrives. This means that your workload can’t be processed right away.
If you are using WLM managed initiators today, this solution will help you intelligently predict upcoming batch workloads and react accordingly for optimized system resources. Based on intelligent automation, your workload spikes can be forecasted and processed right away. They don’t need to wait for resources, and you don’t need to go thru fine tuning and trial and error approaches anymore. The sweet spot is that you can leverage the new AI capabilities infused into WLM without any need for additional AI or data science skills as everything is already pre-built and incorporated for you.
Using the intuitive AI control interface for IBM z/OS, you are able to train your system and switch between AI, non-AI and simulation modes on service class level.
The AI Control Interface for IBM z/OS has a place holder for other use cases. Even if you are not leveraging WLM managed initiators today, you might want to have a look to get acquainted with AI in z/OS and prepared for the future.